REVISION 2 — EVIDENCE REVIEW AND PROSE REVIEW
18 October 2026

EDITORIAL DIRECTION
The initial article spent too much space explaining what it was not claiming and used a generic model-comparison structure. Its strongest pressure-field result was narrow and relied on a single minimum. The revision investigates wind exposure near Manzanillo using finer Met Office fields, the complete available cadence and a later initialization. The public article is now organized around quantified forecast differences and revisions. The original pressure diagnostic remains in the research archive for provenance; it is not the headline finding of the revised article.

NEW DATA
Seven additional MCP requests: one Met Office point probe; four regional pressure/wind requests (GFS, IFS, AIFS, Met Office) initialized 16 October 00 UTC; two bounded coastal-wind requests for the next day's 00 UTC initialization. All returned original CSVs and exact requests are retained. Main window: 17 October 12 UTC through 18 October 06 UTC, inclusive. Cadences: 19 GFS hours, 19 Met Office hours, seven IFS 3-hourly times, four AIFS 6-hourly times. The animation uses GFS and Met Office only, each at its actual hourly times.

SPATIAL RESOLUTION
Met Office catalog native grid: first latitude -89.953125, first longitude -179.9296875, latitude increment 0.09375 degrees, longitude increment 0.140625 degrees. The 4,275 requested coordinates match this served grid within the 15–22 N, 110–102 W box. The API prints coordinates to four decimals, causing only meter-scale rounding. This is the nominal 10 km model's output grid, with approximately 10.4 km north-south and 14.8 km east-west spacing at this latitude; do not advertise all cells as 10 by 10 km. GFS, IFS and AIFS are evaluated on their genuine 0.25-degree served grid. Original cells are retained in the maps; contours use standard linear interpolation across native cell edges, without introducing new weather samples.

CORE FINDING: MATCHED RUN AND VALID TIME
Initializations: all 16 October 2026 00 UTC. Valid time: 18 October 2026 00 UTC, a 48-hour lead; local time is 17 October 18:00 CST (UTC-6). All wind numbers are instantaneous model wind at 10 m, converted by multiplying m/s by 3.6. No gust or NHC sustained-wind intensity is mixed into this metric.

IFS storm-area max 33.8703 m/s (121.93 km/h), Met Office 34.125 m/s (122.85 km/h): near agreement. Within 50 km of Manzanillo (19.05 N, 104.32 W), maxima are IFS 30.9055 m/s (111.26 km/h), Met Office 16.0625 m/s (57.83 km/h). Difference 53.43 km/h. GFS 83.25 km/h and AIFS 60.78 km/h provide additional context rather than hiding the intermediate models. The circle includes offshore water and is not a forecast for a single city location.

Pressure minimum distances from Manzanillo: IFS 78.93 km, Met Office 193.53 km. Their minima are about 115 km different in radial distance from the city; this is not a measured 115 km cross-track error. Native-grid minima: IFS 988.02 hPa, Met Office 982.25 hPa. On 0.25-degree sampling the Met Office minimum is about 983.67 hPa, so the deeper-low contrast remains, though its magnitude changes. This case demonstrates different spatial wind footprints in the forecasts. It does not experimentally isolate track position from storm structure, surface drag, terrain or native-resolution effects as causal mechanisms.

ADVERSARIAL TESTS
1. Place and time: use matching initializations and valid times, not stitched best-available values. The main contrast compares instantaneous regional maxima at one common valid time. The entire surrounding 18-hour sequence is shown to avoid hiding timing differences. The figure explicitly labels native time cadences.
2. Radius sensitivity at 00 UTC: 25 km gives IFS 23.4802 m/s vs Met Office 13.3125; 50 km gives 30.9055 vs 16.0625; 100 km gives 33.8703 vs 20.1875. The ordering is stable. Results cannot be generalized to a different port or radius without recalculation.
3. Common sampling: nearest native Met Office values sampled at the same 0.25-degree point centers give maxima 12.75, 15.5 and 19.5 m/s in the 25, 50 and 100 km circles, respectively. IFS remains higher. This reduces the possibility that a denser point mesh alone created the maximum-wind ordering. It does not equalize physical model resolution.
4. Single-outlier sensitivity: at 50 km, there are 10 IFS grid points and 52 Met Office grid points. Linear-interpolated spatial 95th percentiles are 27.5641 m/s (99.23 km/h) and 15.175 m/s (54.63 km/h). Ordering survives replacing the maximum. Latitude area weighting would differ by less than about a percent over this small circle; the metric is explicitly a grid-point statistic.
5. Nearest-city-point hypothesis: point sampling initially suggested a larger ratio. That was rejected as the headline because different coastline cells and surface exposure could dominate. The article uses a spatial circle and reports scale tests instead.
6. Later initialization: use 17 October 00 UTC runs for the same 18 October 00 UTC valid time and circle. IFS becomes 20.5553 m/s, approximately 74.00 km/h; Met Office becomes 16.25 m/s, 58.50 km/h. Difference 15.50 km/h. The large original difference is strongly cycle-dependent and must not be framed as a persistent model characteristic. The article gives the narrowing its own section and chart.
7. Original pressure-position winner: the earlier AIFS/IFS ordering was sensitive to coarse sampling at the longest lead. The revised article avoids a general winner claim and focuses on forecast disagreement, where the requested data support a much clearer story. The original data and uncertainty notes remain in the package.
8. Tracker stability: the weak GFS pressure field can jump between minima near the end. Regional/coastal wind maxima do not require that tracker. Pressure-minimum paths on maps are labeled as diagnostic paths. No landfall-time claim is made.
9. Temporal interpolation: GFS and Met Office frames are genuinely hourly. IFS and AIFS curves display their actual 3-hourly and 6-hourly output points, connected for readability. Their fields are not invented for intermediate animation hours. The later-cycle evaluation is likewise performed at common real output times.
10. Observational outcome: coastal wind measurements have not been retrieved. Do not say IFS overpredicted, Met Office won, or 111 km/h occurred in Manzanillo. This is a forecast-disagreement and forecast-revision article. The public methods state this limit clearly. The older NHC pressure reference is not used to validate coastal wind.

WHAT THE DATA SUPPORT
The original two models have nearly equal regional wind maxima but very different coastal-area maxima. The spatial footprints and pressure-low positions differ. Much of the wind disagreement narrows by the next day's run. Hourly wind fields expose the rapid fading of the October 16 GFS wind forecast, rather than relying on three pressure snapshots. These are specific, reproducible findings.

WHAT WAS REJECTED
An absolute model winner; extrapolating the forecast difference to observed port conditions; calling grid maxima gusts; treating a pressure-minimum path as an operational cyclone track; describing nominal 10 km output as square 10 km cells; implying finer grids guarantee accuracy; asserting land interaction caused GFS weakening; inventing intermediate IFS/AIFS hours; selecting a single coastal cell to maximize the ratio; framing the 53 km/h split as stable across initialization cycles.

PROSE PASS 1
Removed the old generic headings, abstract 'not this, but that' framing, long opening qualifications, model-league-table narration and repeated statements about unmeasured outcomes. Replaced them with event-specific sections and new geographic/time evidence. Essential metric definitions are next to their first use; full resolution, cadence and verification details are in expandable methods.

PROSE PASS 2
Removed 'the map puts the difference in view' because it narrated the graphic without adding a fact. Replaced the generic coastal-risk paragraph with the concrete consequence for an IFS-only versus Met Office-only dashboard. Added GFS/AIFS values to contextualize the main pair. Checked every paragraph for a distinct job: main contrast, time/place definition, spatial explanation, hourly evolution, later-cycle revision, scale tests or calculation details. The article is approximately half the original length before methods.

MEANING CHECK
The 111 versus 58 km/h numbers refer to forecast maxima across a 50 km region. They are not two weather-station predictions. The 194 versus 79 km numbers refer to distances between sampled forecast pressure minima and Manzanillo. They are not model track errors. The 48-hour and 24-hour comparisons use the same valid time. Every rounded headline number was checked against the retained CSVs.

VISUAL REVIEW
Four-model wind maps share a common color scale and actual valid time. Met Office cell detail is retained, making the compact wind structure visible. Coastal circles and computed regional maxima are labeled. A two-panel hourly animation uses the 19 actual GFS and Met Office output times. A two-panel wind-evolution figure separates the storm-area maximum from the coastal-circle maximum. A revision figure compares the two initialization dates at one fixed checkpoint. Rendered outputs were viewed. A missing arrow glyph was replaced with plain text. GIF first, middle and last frames are checked before delivery. Sources and Met Office attribution are retained in all plots.

PRIMARY CONTEXT SOURCES
NHC Discussion 12, 16 October 2026 at 09 UTC, coast-parallel track sensitivity:
https://prod-east-nhc.woc.noaa.gov/archive/2026/ep21/ep212026.discus.012.shtml
NHC Discussion 3, 14 October 2026 at 03 UTC, earlier track spread and initialization issues (research context, not evidence of the October 16 run's cause):
https://prod-west-nhc.woc.noaa.gov/archive/2026/ep21/ep212026.discus.003.shtml
Met Office global data specification, January 2026 updates and output cadence:
https://www.metoffice.gov.uk/api/assets/file/global-10km-ps47-asdi-pdf-updates-20pdf?prefix=assets
Met Office dataset metadata and exact native geometry retrieved through GribStream MCP.

SOCIAL AND PUBLICATION
Social drafts were rewritten to reflect the wind findings and the next-day narrowing. Existing first browser-attempt permission checkpoint remains active. No composer opened, no publication or outreach attempted. English remains under review.

CONTINUITY
The recurring task was updated to require separate evidence/prose reviews, targeted follow-up of surprising findings, real finer spatial/native temporal data when useful, and removal of empty formulaic prose before delivery.

PROSE PASS 3 — AFFIRMATIVE CLAIMS
Recast the coastal metric as the strongest forecast wind across sea and land within the circle. The conclusion now states the IFS footprint across all tested radii and both wind metrics, followed by its next-day reduction. Methods define the analysis as forecast disagreement and revision and specify the observations required for accuracy assessment. Social copy now states the forecast scope directly and focuses on the changing coastal wind footprint. Figures, numerical results, interpretation strength and publication checkpoints are retained. The recurring editorial instructions now apply this affirmative wording standard to articles, social copy and delivery messages.

PROSE PASS 4 — FINDINGS THROUGHOUT
Applied the findings-led standard across article headings, body, figure captions, chart footnotes, accessible image descriptions, SEO description and all three social drafts. Removed praise for ordinary data handling and repeated announcements of graphics. Recast sensitivity checks as numeric outcomes. Retained exact time, variable, geography and resolution definitions in methods and chart labels. The research log remains in the reproducibility package; the review page leads with the article and social drafts. The recurring task now applies this standard to every reader-facing element and the delivery message. Evidence and forecasts are unchanged.

DELIVERY CLEANUP — REVISION 5
The prior package included a superseded article and social draft alongside the current versions. The package now includes an explicit set of current article, figure, method, code and forecast-source files. The current review uses a fresh filename: trudy-manzanillo-wind-review.html. Original raw forecast CSVs for all six requests remain included. The earlier draft remains recoverable through version history.
